Advanced machine learning approach to increase diagnostic accuracy in atypical Alzheimer’s disease cases

Background The early diagnosis of Alzheimer’s disease (AD) and other types dementia is essential in clinical practice. Up to 50% patients with any form may remain undiagnosed during their lifetime the AD often be inaccurate. In assessing patients, accurate based on only data structural (MRI or CT) imaging suboptimal a subset atypical presentation. Although amyloid PET CSF biomarkers can provide incremental benefits, numerous obstacles preclude wider use this work, parametric study was performed create machine learning (ML) model accurately diagnose hard-to-detect cases AD, mitigate need for more expensive invasive testing such as spinal tap. Method Data from 173 participants (MCI mild dementia), including 105 (60%) 68 (40%) Non-AD were included our memory clinic. dataset each participant pertinent history, risk factors, scores neuropsychological battery, functional status, MRI volumetric studies. Participants divided into two groups: Group I (easy-to-detect) high probability (subjective objective amnestic presentation) studies (hippocampal medial temporal lobe atrophy). II (hard-to-detect), who did not fit first group confirmed either analysis. distribution by group: 1- Easy detect (43 29 Non-AD),2- Hard (62 39 Non-AD), total three experiments different combinations groups conducted, All Groups (Exp.1.), 2- (Exp.2.), 3- (Exp.3.). all experiments, Leave-One-Out cross-validation technique performed. Result A number 132 features utilized, employing feature selection approach, 13 selected. Our attains highest accuracy 87.27% classification Exp.1., 83.28% Exp.2. 95.80% Exp.3. Conclusion We investigated discrimination using ML. proposed method obtains competitive performance achieves improved results.